Interpretation of PlatON White Paper 2.0: How to Build a Decentralized Privacy AI Network?

Original Title: Interpretation of PlatON White Paper 2.0: Realization of Decentralized Privacy AI Network

Written by: Jason, PlatON Community Enthusiast

Recently, PlatON released the white paper version 2.0 on the website. Based on the previous first edition, the new white paper included suggestions from developers and technology enthusiasts in the global community, and revised some details.

PlatON2.0 is an important strategic deployment of PlatON in the next stage. It will form a decentralized artificial intelligence market through the establishment of a decentralized privacy computing network, thereby realizing a self-organizing and collaborative general artificial intelligence network. In order to enable users to fully understand PlatON2.0, open the road to general artificial intelligence, and find the key to technology, PlatON will open a white paper interpretation topic.

When reading an article, the first thing to look at is its structure, not its specific content.

From the structural point of view, the structure of the first three chapters of the article is progressive, and the next few chapters are parallel, and the two parts are connected in the form of total scores.

First, the background and trends are discussed, and the essential requirements are derived from the analysis of observable and perceivable phenomena, and then the inevitability and rationality of the proposed PlatON are obtained. Of course, it is to solve problems according to the requirements. This is the significance of the first three chapters.

Then, the following chapters will describe the deployment or implementation of PlatON from the technical level, application level, economic model, social activity level, plan and progress level. Finally, according to industry practice, the references are given. On the one hand, it shows respect for the intellectual property rights of the authors of the cited articles. On the other hand, it is convenient for friends who have doubts about the cited arguments to trace back and check.

Next, we analyze this context.

overall background

At the beginning of the article, we first pointed out that the era we live in is an era of data explosion, which we all empathize with. The authoritative data statistics and predictions are given in this paper, from Statista and IDC (International Data Corporation), two well-known data analysis institutions.

Generally speaking, it expresses a point of view, or phenomenon, that the amount of data will increase exponentially in the future, and the new data born with new technologies and applications such as 5G and the Internet of Things has great commercial significance. Value is very much in need of protection. However, if the current data management model continues, by 2025, nearly half of the data that should be protected will not be protected.

This is a very serious phenomenon, and it is also a pain point for the collective advancement of the industry in the future. Of course, from another perspective, this is also a very important opportunity.

artificial intelligence and data

From the perspective of the overall economic and industrial development trends, the importance of data privacy protection is certainly reasonable, but the resulting problem is that it is easy to fall into ambiguity.

Indeed, from a broad perspective, data privacy protection, as a basic function, can be said to be required and applicable to all applications in the digital economy era. It is necessary to find a focused entry point for privacy computing to realize the first application Concentrated presentation of breakthroughs and effects.

AI, artificial intelligence, as a hot spot in recent years, has its manifestations in both policy industry planning and industry applications.

Also, and importantly, AI does have an incessant and unreasonable connection to data. Artificial intelligence has always had a development idea based on logic and machine learning. Early AI was mainly based on logic, such as expert systems, knowledge graphs, etc. The core was to compile rules, which mainly relied on human experience, and the algorithm itself did not require much data.

Since the beginning of the 21st century, especially in recent years, thanks to the convenience of data collection and storage, the improvement of computer performance, and the improvement of software processing architecture, AI based on machine learning has begun to reappear, and it has been out of control. Among them, the well-known representatives should be heard by everyone. AlphaGo, which defeated a lot of Go masters, is behind deep learning.

The core idea of machine learning is to enable the machine to continuously learn through relevant data. If the answer to the test after learning is correct, then strengthen the memory. If the answer to the test after learning is wrong, then correct the learning method until able to answer correctly.

In fact, it is essentially the same as our reading and learning.

We start reading and studying in kindergarten, and we read more and more books, and the books become more and more difficult. At the same time, our knowledge and cognitive level are gradually improving.

The same is true for artificial intelligence based on machine learning. We hope that the higher the level of artificial intelligence, we need to find more categories, larger quantities, and better quality data for training.

At present, in fact, many companies' AI itself has little gap in model architecture and algorithm design. What really widens the gap is its training level, or in other words, how much data can be used for the training of its AI model.

Problems and pain points

As mentioned earlier, current artificial intelligence requires a large amount of high-quality data of different types for training, which is the key to effectively improving the level of intelligence, and its importance often exceeds the design of the algorithm itself. But there is a problem, that is, where does the data come from?

Internet giants have a natural advantage because they have their own platforms and collect user data all the time, and these companies are still expanding their data territory through a large number of acquisitions and investments.

For us ordinary people, because we have long been accustomed to various seemingly free business models, we don’t feel deeply. In fact, this has brought about three deep-seated problems.

The first is the AI monopoly problem brought about by the data monopoly phenomenon. The Internet giants naturally have data resources, so it is easier to develop their own AI, and their AI also naturally has a higher level of intelligence, which causes the Matthew effect in the AI field. In the end, the entire AI will also be controlled by these Internet giants, repeating the current situation. The mistakes of the mobile Internet era will not only be the pricing power of the public's food, clothing, housing and transportation, but also our decision-making and thinking will be deeply affected by these giants.

The second is the AI ceiling problem brought about by data privacy concerns. For ordinary people, most of us choose to trade data for convenience due to the lack of systematic management capabilities for personal data. This loss is hidden and usually not obvious, but for enterprises, other A series of data such as scientific research, production, operation, personnel and finance are actually the most important business secrets, and few companies are willing to share their core data.

Therefore, we often see that there are data islands among various departments of the enterprise and between enterprises. Even among Internet giants, there will be mutual boundaries, which will lead to a large amount of cross-category high-quality data being frozen for a long time and cannot be used for AI training. This must be a critical point that restricts the future development of AI.

Finally, there are issues of data ownership, use rights, and usufruct rights. As mentioned before, each person's data does not seem to be of much value, but in the hands of various Internet giants, it is through our data that a series of behaviors such as financing and realization have been realized. That is to say, our data is actually valuable , but today, due to various objective constraints and subjective manipulations, these values have been blurred. Then, in the era of data explosion in the future, when the value of data can no longer be ignored, if this value can be clarified and guaranteed Series rights, this is also a problem.

solution

It is not difficult to see from the PlatON 2.0 white paper that the vision of PlatON is to build a decentralized collaborative artificial intelligence network and global brain. Its core idea is to capitalize data, so as to realize the free circulation of data, and then greatly accelerate the maturity and application of AI algorithms.

This is a systematic solution to the aforementioned problems.

It can be seen that the first step is to realize the capitalization of data, which is the foundation. There is no doubt that data itself is a valuable factor of production, but being valuable does not mean that it can be capitalized. There are two keys to capitalization, one is right confirmation, and the other is protection. That is, how to establish the ownership of data, and by what means to protect sovereignty.

Ordinary assets can achieve the above goals through legal means, but for intangible assets such as data, it is more difficult to pass laws alone, because the copying and dissemination of data is extremely simple, and it is troublesome to pursue accountability after the event, not to mention the prior event. protected.

Therefore, it has become a mainstream perception to confirm and protect rights through technical means, and PlatON is realized based on blockchain + privacy computing technology. Among them, privacy computing technology is the key to realizing data rights confirmation and protection, and the role of blockchain is to facilitate the scheduling of resources and the circulation of data.

Due to the different functions of the technology stack, for the purpose of decoupling, from the architecture, PlatON is decomposed into three layers, each performing its duties.

The first layer is the consensus layer, where the blockchain technology stack is located, including nodes, consensus mechanisms, and smart contract virtual machines;

The second layer is the privacy computing network, which is where the privacy computing technology is located, including the algorithms and data that the implementation of the privacy technology relies on, as well as the computing nodes that deploy and execute the algorithms and the data nodes that provide data. The algorithms are all based on cryptography technology. Including secure multi-party computation, homomorphic encryption, zero-knowledge proof, etc.;

The third layer is the collaborative AI network. To put it bluntly, it is a shelf of AI models. These AI models can be trained based on the previous two layers. This shelf can be continuously updated.

features

PlatON is positioned as a private AI computing network, and its core is private computing. However, at the same time, it should be noted that it is also a Turing-complete public chain platform. That is to say, from a functional point of view, it is similar to Ethereum. , supports virtual machines and smart contracts. Being functionally similar means that they will face similar issues in implementation.

This is a problem that Ethereum has been unable to avoid for a long time. It is also a problem that has become more and more uncontrollable after the explosion of various applications such as DeFi and NFT in the past two years. It is the problem of TPS, or scalability.

Ethereum 2.0 will use fragmentation to improve this problem. Currently, Ethereum 1.0 uses layer2 to improve this problem.

Specifically, PlatON adopts a layered approach. Strictly speaking, it is also a layer2 solution. The core is still through the layered architecture mentioned above. While decoupling is achieved, it also greatly improves the performance of the entire network. Horizontal scalability. The on-chain and the off-chain are linked by Verifiable Computing (VC) based on cryptography, that is, specific personalized computing tasks are performed off-chain, and the completion of the task is verified on the chain and a consensus is reached across the network. Easy-to-extend off-chain, reducing the occupation of scarce on-chain resources.

Back to the core point, privacy computing. In fact, privacy computing is not a new technology. Research on this aspect has been going on for years. Looking at the current privacy computing related projects in the industry, the technical paths can generally be divided into two categories , one is based on TEE (Trusted Execution Environment) hardware, among which the well-known TEE is Intel's SGX; .

The advantage of TEE-based privacy computing implementation is that it is easy to implement in engineering, but the disadvantage is that it brings centralized dependence on hardware manufacturers, and is limited by the performance of TEE itself, which has problems in scalability; privacy computing based on cryptography technology The advantage of implementation is that it is more secure, there is no centralization dependence, and the scalability space is higher. The disadvantage is that with the current technical level, the project implementation is not ideal, and it is difficult to achieve both efficiency and versatility. Taking the path of cryptography requires a sufficient scientific research foundation and continuous funding.

Above, the characteristics of PlatON2.0 have been sorted out. A brief summary of the content in the white paper is:

• Support WASM and EVM virtual machines, good compatibility with Ethereum;
• Layered system architecture decoupling on-chain and off-chain, with strong system scalability;
• A privacy protection solution based on cryptography technology with a high upper limit.

peripheral tools

The concept of private computing is very easy to understand, and its utility is easy to accept.

However, this does not change the hidden fact that the technical threshold of privacy computing is actually quite high. For ordinary AI practitioners, if they want to achieve AI model training through privacy computing, they need to have AI-related In addition to knowledge, you also need to have high attainments in cryptography to integrate the privacy framework into the AI model well, which is actually very unfavorable for the promotion of privacy computing applications.

So here is a separate mention of Rosetta, the privacy AI development framework in the white paper. In my opinion, Rosetta is very important. Although it can actually exist independently of the PlatON network, it does not seem to have much relationship, but it is actually very important for accelerating In terms of the growth of the entire system, it is very important, so it is necessary to talk about it separately.

To put it simply, Rosetta is actually a big data processing framework, such as TensorFlow, etc.; commonly used AI algorithms, such as statistical machine learning, deep learning based on artificial neural networks; cryptographic algorithms, such as secure multi-party computing, homomorphic encryption; and The underlying hardware scheduling framework is packaged and integrated together, with internal fusion processing, and only some commonly used calling interfaces are exposed to the outside, so that for AI engineers, when developing AI algorithms, they only need to consider the logic of their own business , and call the corresponding interface to realize it, without having to learn more complex cryptographic algorithms.

In the future, developers can use the second-tier privacy computing network of the system to conveniently develop and train their own AI models through Rosetta, and upload them to the third-tier collaborative AI network for use in specific applications.

Application Scenario

As mentioned before, the entire digital economy system needs the support of data privacy protection, because in the digital economy, data is a valuable, or more directly, priced asset. Therefore, through privacy protection, data assets Affirmation of rights and protection is very necessary.

The white paper lists several typical application scenarios, including hotspots related to the blockchain industry, such as oracle machines and blockchain games, as well as traditional biomedicine, financial risk control, and smart cities.

So in which part of these application scenarios will privacy computing be used? In fact, without exception, it is in the link that requires the interaction of multiple parties.

There are some cases mentioned in the white paper, so I won’t repeat them. I will try to describe them with familiar situations.

For example, for the oracle machine, the oracle machine is the source of specific information for users on the chain. However, for users, the act of continuing to pay attention to and obtain those data itself is a kind of privacy in itself, and may even involve business secrets. At this time Privacy protection is very important.

For example, for chain games, everyone knows that one direction of chain games is metaverse. Metaverse tries to build a realistic and immersive world. Restoration in the virtual world, if there is no privacy computing to confirm and protect the data, the data owners in the real world may not be willing to provide their truly valuable data.

For example, for biomedicine, an important part of the progress of medical care comes from the analysis of a large number of clinical cases. Usually, large hospitals have a lot of data in this area, while many small and medium hospitals are relatively scarce. From the perspective of patient privacy and their own development, hospitals usually They are unwilling to share these data, because of the particularity of the data, once shared, it will be out of control, which largely restricts the development of the overall level of medical care, and after the confirmation and protection of data based on privacy computing , it is possible to solve this problem;

For financial risk control, in theory, the more comprehensive the information you have about the risk rating of a person or company, the more accurate the assessment result will be. There are 5 aspects of compliance record.

So here comes the problem. Ali may not be able to have more comprehensive data on people connections than Tencent, and data such as assets and behaviors may also be available on other platforms besides Ali’s platform. At present, the data between these Internet platforms is actually disconnected, so the accuracy and versatility of user portraits may be lacking. For example, high credit on one platform may not be directly accepted by another platform.

If privacy calculations can be used to comprehensively utilize data from multiple platforms without compromising the data privacy of all parties, a more accurate and universal portrait of people can be obtained, making it easier to provide risk matching information in a targeted manner. Financial Services.

For smart cities, one of the purposes of building a smart city is to make urban services more convenient, more accurate, more efficient, and safer. In order to achieve this goal, it often requires the cooperation of various government departments and some enterprises. Taking urban security as an example, this requires the linkage of relevant departments’ Skynet cameras, road traffic cameras, residential property cameras, metal detection sensors of various commercial institutions, and communication data of operators, so that suspicious and dangerous situations can be detected immediately And continuous tracking, prevent problems before they happen, and these sensors belong to different subjects, it is necessary to protect the privacy of these sensor data no matter for the protection of citizens' privacy or for the consideration of various subjects.

epilogue

Dreams are necessary, and results are also necessary. For me, what I am most looking forward to is the launch of the private computing network at the end of this year. In my opinion, this is the core competitiveness of PlatON.

About here, the privacy AI landscape of PlatON should be relatively clear. For more specific situations, you can also read the white paper in detail.

Finally, I want to say that what I am talking about here is the white paper of PlatON 2.0, and it is the overall track of privacy. This track is very difficult. Who can really make it to the end? To be honest, I don’t know. This requires capital The joint promotion of technology, technology, and business also requires a lot of patience and perseverance, but in any case, I believe this is the right direction, and eventually someone will go down.

Source: Lianwen

Disclaimer: Cointelegraph Chinese is a blockchain news information platform, and the information provided only represents the author's personal opinion, which has nothing to do with the position of the Cointelegraph Chinese platform, and does not constitute any investment and financial advice. Readers are requested to establish correct currency concepts and investment concepts, and earnestly raise risk awareness.